Fast TLS Denoising Algorithm Using Grid Technique

The Total Least Squares (TLS) is a denoising approach to take into the small system perturbations, and to restore the noisy images through a linear combination of the image patches. It can handle not only additive noise but also mixture noise. However, it suffers from high computational complexity and low efficiency. To this end, this paper proposes a grid-based denoising algorithm to speed up the TLS algorithm. Firstly, we utilize a mean value method instead of the bilateral filter parameters used in original algorithm to calculate the weight matrix, and then a strategy of making calculation every half patch is applied. It is proved by experiments that the proposed image denoising algorithm is capable of increasing the calculation speed by two orders of magnitude.